test(chat): tear the coordinator down however a test ends

[bmc08gt]

A coordinator test that failed its assertion hung the run instead of reporting the failure.

RealChatCoordinator.onUserLoggedIn ends with startHeartbeat { syncFeeds() }, a while (true) of delays on the coordinator's own CoroutineScope. That scope is not the test's backgroundScope, so it outlives the test body, and runTest drains the scheduler once the body returns. These tests called coordinator.teardown() as a bare last statement, so an assertion that threw skipped it and left that loop advancing virtual time with nothing to stop it.

Each body now runs inside a tornDown helper that tears down in a finally.

Evidence

Same deliberate break (exactly = 1 to exactly = 99) in ViewerStateStreamTest, run twice:

Arrangement	Result
Before this change	still spinning when killed at 240s; no failure reported
After this change	fails in 14s, naming all three affected tests

Module suite green at 13s.

Scope

25 tests across four files. GroupChatRoutingTest and ChatHistoryHydrationTest already used this shape and document the hazard in KDoc — this brings the rest onto the same footing rather than inventing a pattern.

ChatCoordinatorTeardownTest is deliberately untouched: it calls teardown() as the subject under test and never logs in, so no heartbeat is running. The one deliberate mid-body teardown() in ChatCoordinatorEventsTest's re-login test is also preserved — only its trailing call became the finally.

Tests only; no production code.

Why separately from #1511

Found while debugging a hang in a new test on that branch. It is a pre-existing hazard in tests unrelated to that fix, so it reviews better on its own.